incubator-cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jason Wee <peich...@gmail.com>
Subject Re: Write performance with 1.2.12
Date Fri, 06 Dec 2013 15:58:46 GMT
Hi srmore,

Perhaps if you use jconsole and connect to the jvm using jmx. Then uner
MBeans tab, start inspecting the GC metrics.

/Jason


On Fri, Dec 6, 2013 at 11:40 PM, srmore <comomore@gmail.com> wrote:

>
>
>
> On Fri, Dec 6, 2013 at 9:32 AM, Vicky Kak <vicky.kak@gmail.com> wrote:
>
>> Hard to say much without knowing about the cassandra configurations.
>>
>
> The cassandra configuration is
> -Xms8G
> -Xmx8G
> -Xmn800m
> -XX:+UseParNewGC
> -XX:+UseConcMarkSweepGC
> -XX:+CMSParallelRemarkEnabled
> -XX:SurvivorRatio=4
> -XX:MaxTenuringThreshold=2
> -XX:CMSInitiatingOccupancyFraction=75
> -XX:+UseCMSInitiatingOccupancyOnly
>
>
>
>> Yes compactions/GC's could skipe the CPU, I had similar behavior with my
>> setup.
>>
>
> Were you able to get around it ?
>
>
>>
>> -VK
>>
>>
>> On Fri, Dec 6, 2013 at 7:40 PM, srmore <comomore@gmail.com> wrote:
>>
>>> We have a 3 node cluster running cassandra 1.2.12, they are pretty big
>>> machines 64G ram with 16 cores, cassandra heap is 8G.
>>>
>>> The interesting observation is that, when I send traffic to one node its
>>> performance is 2x more than when I send traffic to all the nodes. We ran
>>> 1.0.11 on the same box and we observed a slight dip but not half as seen
>>> with 1.2.12. In both the cases we were writing with LOCAL_QUORUM. Changing
>>> CL to ONE make a slight improvement but not much.
>>>
>>> The read_Repair_chance is 0.1. We see some compactions running.
>>>
>>> following is my iostat -x output, sda is the ssd (for commit log) and
>>> sdb is the spinner.
>>>
>>> avg-cpu:  %user   %nice %system %iowait  %steal   %idle
>>>           66.46    0.00    8.95    0.01    0.00   24.58
>>>
>>> Device:         rrqm/s   wrqm/s   r/s   w/s   rsec/s   wsec/s avgrq-sz
>>> avgqu-sz   await  svctm  %util
>>> sda               0.00    27.60  0.00  4.40     0.00   256.00
>>> 58.18     0.01    2.55   1.32   0.58
>>> sda1              0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> sda2              0.00    27.60  0.00  4.40     0.00   256.00
>>> 58.18     0.01    2.55   1.32   0.58
>>> sdb               0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> sdb1              0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> dm-0              0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> dm-1              0.00     0.00  0.00  0.60     0.00     4.80
>>> 8.00     0.00    5.33   2.67   0.16
>>> dm-2              0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> dm-3              0.00     0.00  0.00 24.80     0.00   198.40
>>> 8.00     0.24    9.80   0.13   0.32
>>> dm-4              0.00     0.00  0.00  6.60     0.00    52.80
>>> 8.00     0.01    1.36   0.55   0.36
>>> dm-5              0.00     0.00  0.00  0.00     0.00     0.00
>>> 0.00     0.00    0.00   0.00   0.00
>>> dm-6              0.00     0.00  0.00 24.80     0.00   198.40
>>> 8.00     0.29   11.60   0.13   0.32
>>>
>>>
>>>
>>> I can see I am cpu bound here but couldn't figure out exactly what is
>>> causing it, is this caused by GC or Compaction ? I am thinking it is
>>> compaction, I see a lot of context switches and interrupts in my vmstat
>>> output.
>>>
>>> I don't see GC activity in the logs but see some compaction activity.
>>> Has anyone seen this ? or know what can be done to free up the CPU.
>>>
>>> Thanks,
>>> Sandeep
>>>
>>>
>>>
>>
>

Mime
View raw message